whittle 0.2.0

Two-stage neural architecture search for large language models

whittle

A framework for two-stage neural architecture search (NAS) and structural pruning on language models.

Setup

1. clone this repository: git clone [e-mail]:whittle-org/whittle.git
2. cd into the repository: cd whittle
3. Create a new conda environment and activate: conda create -n whittle python=3.11 -y && conda activate whittle
4. Install the dependencies: pip install .

Note: the library supports and is tested for python 3.9 to 3.11